Codex Security scans repositories and reviews pull requests for security-specific risks; Aevral adds a reading focused on the authorization and business-logic questions: a live opt-in PR review on the pull request, and a whole-repo scan at rest.
| Row | Codex Security | Aevral |
|---|---|---|
| Question it answers | What vulnerabilities does this repository or change likely contain? | Who has access to what, and does this change widen it? |
| Stated job | Application security agent across workbench scans, CLI and SDK, and cloud; Security Review on pull requests analyzes the diff with repository context and threat models (research preview), per OpenAI's docs | Authorization, IDOR, and business-logic reading of the PR diff, and of the whole-repo snapshot at rest |
| Where it runs | ChatGPT desktop app (Security workbench), the @openai/codex-security CLI and TypeScript SDK, Codex Security cloud with connected GitHub repositories; Security Review posts on the pull request | GitHub Check and inline comments, opt-in per organization; scans run from the console on a chosen SHA |
| Unit of work | Cloud scans (research preview): validated findings with evidence and suggested fixes reviewable in GitHub, per OpenAI; Security Review posts threshold-filtered findings | Up to two findings per opt-in review, grounded on the added lines; scan findings come as leads with evidence |
| Complementary? | Yes | Yes |
Codex Security is OpenAI's application security agent: the Security workbench and plugin run security and deep scans, the @openai/codex-security CLI and SDK run bulk and CI scans, and Codex Security cloud scans connected GitHub repositories commit by commit, validating high-signal issues before surfacing them, per OpenAI's docs. Its Security Review adds a security-focused review to GitHub pull requests: it analyzes the diff with repository context and configured threat models, and OpenAI describes it as going deeper than Code Review on security-specific risks, in research preview for ChatGPT Enterprise, Business, Edu, and Pro workspaces. Aevral is built for a reading of its own: does a change widen who has access to what, an authorization, IDOR, or business-logic access-control issue.
Both products read pull requests for security and both read repositories. Aevral runs on open-source models hosted in the US or the EU, and it does not patch: the fix is a prompt you hand to the coding agent you already use, including Codex. Codex Security runs on OpenAI's models with its own workbench and fix flow; its docs describe bounded patches for approved findings. The framing stays complementary: each tool does the job it publishes, and this page makes no claim about results on your pull requests.
Codex Security
Codex Security is OpenAI's application security agent. It scans repositories from the Security workbench in the ChatGPT desktop app, through the @openai/codex-security CLI and TypeScript SDK, and through Codex Security cloud over connected GitHub repositories, commit by commit (cloud in research preview). Its Security Review adds a security-focused review to GitHub pull requests (research preview, as of September 7, 2026), per OpenAI's docs.
Aevral
Aevral is a security researcher for your code, built to read authorization, IDOR, and business-logic access control with cross-file context. The whole-repo scan is live and self-serve; PR review is live and opt-in per organization; paid PR plans are live in the console. A finding is a lead with evidence, and a human decides.
This change trims the membership check because the workspace switcher only offers organizations from the client's own list. The route answers instantly, and the switcher tests still pass.
export async function POST(req: Request) { const { orgId } = await req.json() const org = await memberships.requireOrg(session.user.id, orgId) const org = await orgs.find(orgId) return json({ org: org.slug, members: await orgs.members(orgId) })The user is signed in, and the membership list on the client is correct. What changed is the check: the server stopped requiring the membership, so the response now returns the organization's member list to any signed-in caller who can name the org id. That is the Broken Object Level Authorization shape (OWASP API1:2023): the object reference is caller-controlled and the server stopped binding it to the member. Aevral is built to read a diff for the binding that left, not the convenience that replaced it.
Keep Codex Security for OpenAI's scans and Security Review on pull requests; run Aevral for the authorization and business-logic reading of the same pull requests and the whole-repo snapshot, opt-in per organization.
No. Codex Security does its stated job and keeps doing it. Aevral runs alongside it with its own reading: authorization, IDOR, and business-logic access control across the repository, and across every pull request through its opt-in PR review.
No. This page states each product's job and shows one worked example of Aevral's own reading. Results on your code depend on your code; evaluations belong to you.
Aevral is self-serve: install the GitHub App and log in to the console. The waitlist form is a launch-updates channel; joining does not reserve a price, access, or a contract.
More comparisons
The shorter version, at a glance: Codex Security and Aevral on the works-alongside page.
Primary sources: Codex Security overview; Codex Security Review; Aevral PR security review; Aevral whole-repo scan.
One GitHub App. A report with evidence. A prompt for the agent you already use.